Browser-based simulators that turn equity-analytics and decision cases into things you can move a slider on — each launchable in one click.
The conceptual home base for the course. Presents the signature 2×2 — Process (Allocations × Valuations) crossed with Behavior (Differential Treatment × Disparate Impact) — as a clickable matrix, plus the seven-step analytics workflow and the recurring maxims that every case simulator plugs into.
A $200M cost cut at a 30,598-employee auto-parts firm. Pick among seven identity-neutral downsizing criteria and watch each one's adverse impact by race and gender: every option hits the target, none violates the four-fifths rule, and each harms a different protected group. An "Option 8" coin flip sets the zero-impact benchmark.
Recidivism risk scores on the ProPublica data (N = 7,214). Slide the classification threshold and toggle group-specific cutoffs to see false-positive and false-negative rates diverge by race despite near-identical AUC — landing on Chouldechova's impossibility theorem: calibration, equal FPR, and equal FNR cannot all hold when base rates differ.
Cutting an admit class from 8,000 to 5,200 under Prop 209 and SFFA v. Harvard. Each identity-neutral removal criterion lands in a 2×2 of mechanism versus who bears the harm — surfacing the convergence question, negative action against Asian admits, and the denominator move that quietly shifts the cost.
A LASSO-penalized hiring model advancing the top 25% of applicants (N = 2,500). Build your own specification, then toggle the outcome variable between "predict hired" and "predict aptitude": identical accuracy (AUC = 0.849) and an identity-blind feature set produce dramatically different racial composition — because choosing the outcome is choosing what the model is for.
A stylized four-unit employer audits pay across 5,000 people and finds no significant gender gap. Disaggregate and two units appear with large gaps pointing opposite ways. A reweighting lab then changes only headcounts — no one's pay — and flips the sign of the headline number. Open to visiting instructors, with the generating parameters published and a link to the related case.
More from Christopher I. Rider →